Skip to content

feat: first reviewed-character video workflow - #9

Draft
nkcbuilds wants to merge 3 commits into
mainfrom
feat/first-video-mvp
Draft

nkcbuilds wants to merge 3 commits into
mainfrom
feat/first-video-mvp

Conversation

@nkcbuilds

Copy link
Copy Markdown
Owner

First usable vertical slice (one reviewed character)

This feature branch provides a narrow, repeatable approved rig -> editable scene -> voice alignment -> H.264/AAC MP4 workflow. It does not promise universal one-click PNG rigging, arbitrary pose bridges, new AI artwork or automatic TTS.

Changes

  • character approve <project> --reviewed: explicit acknowledgement gate for one provenance-tracked imported Rev2D rig, requiring current SHA-256 source/spec/rig/texture integrity and existing review PNGs. Never auto-approves.
  • scene new <scene.json> --character <character.json> [--id .. --seconds .. --fps .. --size .. --voice WAV --cues JSON]: strict semantic scene generation, uses real native rig dimensions, refuses unapproved or unsupported capabilities, audio requires true Rhubarb cues, creates unique filename-based IDs and never overwrites existing JSON.
  • Clear scene validate renderability errors for deterministic unsupported captions, motions, missing lip cues, multi-actor and bridge clips; the broader planning validator API remains intact.
  • Safe exports: FFmpeg writes only to a unique temporary sibling MP4; atomic no-clobber promotion by default, --force explicit replace after successful encoding. Added filesystem realpath checks for CLI scenes, model, audio/cues and output paths.
  • FFprobe acceptance checker, reproducible two-scene MIT Hana smoke script, agent skills, README and first-video guide.

Actual evidence (ignored local files, not committed)

  • Two distinct 5-second Hana scenes rendered from one copied pinned Rev2D test fixture, validated, visually inspected and accepted by FFprobe: 512×640, 15fps, 75 H.264 frames each, no audio, different expression/look events.
  • One 12-second voiced scene from the existing approved upstream Hana rig, using a private local test WAV aligned by installed Rhubarb into 36 true mouth cues. Result: 180 frames, 15fps, H.264 video, AAC audio, exactly one stream of each and correct duration; FFprobe pass. Inspected three real frames. Local sample file: .cache/first-video-demo/qa-12s-voiced/hana-voiced-12s.mp4.
  • Confirmed explicit --force re-render over an existing disposable MP4 succeeds; without force, regression tests verify old bytes survive.
  • npm run verify green: 149 tests / 26 files, typecheck and build. git diff --check clean.

Still blocked / not ready for universal public MVP

  • Single-image imported Hana previously had unacceptable eyelid/face artifacts and stays unapproved; current verified voice/scene proof uses already-rigged upstream licensed art.
  • Natural-speaker perceptual lip sync, voice-cue/source hash binding, diverse artwork/occlusion cases, arbitrary head/pose bridging, captions, TTS, multiple actors and GUI are still open.
  • Draft PRs feat: guarded local drawn eye and mouth variants #7 (drawn variants) and feat: bounded PNG animation quality probes #8 (pixel/temporal QA) are separate and overlap docs/CLI; do not auto-merge until conflicts and quality are reviewed.

@coderabbitai

coderabbitai Bot commented Nov 7, 2026

Copy link
Copy Markdown

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

  • Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

Share

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant